I have constructed several models in R and used a model selection process to select the most appropriate model based on AIC values. However, I am wanting to use AICc for model selection, not AIC. I can calculate AICc values for each model, but was wondering if I also need to convert the AIC weight, cumulative weight and delta values as well so that they are relevant to the new AICc values and not the old AIC values? I provide an example below of the model comparison using AIC. Thanks in advance.
nPars AIC delta AICwt cumltvWt
model1 6 157.25 0.00 9.4e-01 0.94
model3 3 162.72 5.47 6.1e-02 1.00
model2 5 178.70 21.44 2.1e-05 1.00
modelnull 2 180.40 23.15 8.8e-06 1.00
Hi Patrick,
Yes, I would suggest that if you need to use AICc (AIC corrected for small sample sizes), then you should re-rank your models using AICc and then recalculate the delta AICc and cumulative weights accordingly. I am not sure what software you are using, but it is pretty easy to calculate AICc and derive a model selection table using these AICc values in the R package AICcmodavg.
Cheers,
Steph
Thanks Stephanie.
I am using R and can calculate AICc values for each model in the package AICcmodavg, but cannot work out how I calculate the other values (AICc weight, cumlative weight, delta) or compare multiple models at once using the AICc function in AICcmodavg. I provide an example below. I would like to create a table similar to what I have done using modSel for AIC, but instead for AICc (or if creating a table is not possible then just obtaining the other values would be great)
> # model selection -- ranked list of models and AIC
> model_list modSel(model_list)
nPars AIC delta AICwt cumltvWt
model1 6 157.25 0.00 9.4e-01 0.94
model3 3 162.72 5.47 6.1e-02 1.00
model2 5 178.70 21.44 2.1e-05 1.00
nodelnull 2 180.40 23.15 8.8e-06 1.00
>
> # AICc values for each model
> AICc(m4, return.K = FALSE)
[1] 163.2511
> AICc(m5, return.K = FALSE)
[1] 182.696
> AICc(m6, return.K = FALSE)
[1] 164.1331
Hi Patrick,
If you use aictab from the package AICcmodavg - you should be able to get it to rank your models using AICc (and derived model weights). You just supply it a list of models, and a list of model names to it, and insert the command 'second.ord=T' (although this is the default anyway!) - and it will create the table using AICc. For example:
model_list
Thanks Steph, this worked :) I had been playing around trying to work that out for ages!
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Software
- R Statistical Package